{"id":106703,"date":"2025-02-20T07:59:45","date_gmt":"2025-02-20T13:59:45","guid":{"rendered":"https:\/\/engineering.wisc.edu\/?post_type=tribe_events&p=106703"},"modified":"2025-02-26T12:21:51","modified_gmt":"2025-02-26T18:21:51","slug":"engineering-robust-scalable-ai-for-healthcaresystems","status":"publish","type":"tribe_events","link":"https:\/\/engineering.wisc.edu\/event\/engineering-robust-scalable-ai-for-healthcaresystems\/","title":{"rendered":"ISyE Engineering Robust & Scalable AI for Healthcare Systems"},"content":{"rendered":"
Artificial intelligence (AI) is increasingly used in healthcare to enhance clinical decision-making,
optimize operations, and improve patient outcomes. However, real-world deployment of AI
systems presents fundamental engineering challenges, including dataset shifts, physician-AI
team dynamics, and the need for continuous model validation and updating. These challenges
threaten the reliability and scalability of AI tools, limiting their ability to provide consistent value
in clinical environments.<\/p>\n\n\n\n
In this talk, I will present engineering solutions that address these core challenges and enable
the development of AI systems that are both scalable and safe. First, I will discuss techniques
for integrating longitudinal patient data into predictive models, improving their performance
over time. Second, I will introduce methods to detect and mitigate dataset shifts, ensuring
models maintain accuracy when transitioning from development to real-world use. Finally, I will
describe a novel rank-based compatibility measure and optimization framework that improves
model updating while preserving physician trust and workflow stability.<\/p>\n\n\n\n
By developing these foundational methods, my work moves healthcare AI from an artisanal,
model-by-model approach to a scalable engineering discipline. I will conclude by discussing
future research directions, including AI personalization for individual physicians and the
development of interactive AI validation systems that continuously adapt based on clinician
feedback.<\/p>\n\n\n